Fast saccades toward faces: Face detection in just100 ms

Centre de Recherche Cerveau and Cognition, UMR, CNRS,Université Toulouse, Toulouse, FranceSébastien M. Crouzet

Centre de Recherche Cerveau and Cognition, UMR, CNRS,Université Toulouse, Toulouse, FranceHolle Kirchner

Centre de Recherche Cerveau and Cognition, UMR, CNRS,Université Toulouse, Toulouse, FranceSimon J. Thorpe

Previous work has demonstrated that the human visual system can detect animals in complex natural scenes veryefficiently and rapidly. In particular, using a saccadic choice task, H. Kirchner and S. J. Thorpe (2006) found that when twoimages are simultaneously flashed in the left and right visual fields, saccades toward the side with an animal can be initiatedin as little as 120–130 ms. Here we show that saccades toward human faces are even faster, with the earliest reliablesaccades occurring in just 100–110 ms, and mean reaction times of roughly 140 ms. Intriguingly, it appears that these veryfast saccades are not completely under instructional control, because when faces were paired with photographs of vehicles,fast saccades were still biased toward faces even when the subject was targeting vehicles. Finally, we tested whether thesevery fast saccades might only occur in the simple case where the images are presented left and right of fixation by showingthey also occur when the images are presented above and below fixation. Such results impose very serious constraints onthe sorts of processing model that can be invoked and demonstrate that face-selective behavioral responses can begenerated extremely rapidly.

Keywords: fast visual processing, face detection, visual processing time, saccade, eye movement

Citation: Crouzet, S. M., Kirchner, H., & Thorpe, S. J. (2010). Fast saccades toward faces: Face detection in just 100 ms.Journal of Vision, 10(4):16, 1–17, http://journalofvision.org/10/4/16/, doi:10.1167/10.4.16.

Introduction

Measurements of processing speed in the visual systemcan be very useful for constraining models. For example,in a manual go/no-go task, subjects can reliably release abutton when an animal is present in a natural scene fromaround 300 ms after stimulus onset (although meanreaction times are longer), and in the same situation, thereis a differential EEG response between target anddistractor trials that appears only 150 ms after stimulusonset (Fabre-Thorpe, Delorme, Marlot, & Thorpe, 2001;Thorpe, Fize, & Marlot, 1996). These latencies areundoubtedly very short given the computational complexityof the task and have led to the suggestion that at least somesorts of high-level visual tasks can be performed on thebasis of a single feed-forward sweep through the visualsystem (Serre, Oliva, & Poggio, 2007; Thorpe & Imbert,1989; VanRullen & Thorpe, 2002). Nevertheless, there isevidence that processing involving feedback can occurvery rapidly. For example, the question of whether aparticular region of the image is foreground or back-ground affects activity in areas such as V1 within a fewtens of milliseconds of the start of the neural response (Qiu,Sugihara, & von der Heydt, 2007; Roelfsema, Tolboom, &

Khayat, 2007). As a consequence, even processing involv-ing both a feed-forward and a feed-back pass could bepossible very rapidly (Epshtein, Lifshitz, & Ullman, 2008).For this reason, precise measurements of processing timeare likely to become even more important for distin-guishing between processing that can be achieved with asingle feed-forward pass and processing that leaves enoughtime for both bottom-up and top-down mechanisms to beinvolved.While much of the evidence for ultra-rapid scene

processing has come from using a manual go/no-gocategorization task coupled with event-related potentials,it has become clear that there are limitations to thisapproach. One of the strongest arguments in favor of theidea that the differential EEG response at around 150 msis indeed related to categorization comes from the fact thatthe differential activity can be modulated by changing thetarget category (VanRullen & Thorpe, 2001b). Thus,when subjects were required to switch the target categoryfrom “animal” to “means of transport” in different blocks,there were differences in the ERP response from around150 ms that depended not on the physical characteristicsof the stimulus but rather on the status (“target vs.distractor”) of the image. This effectively rules out thepossibility that the differential activity simply results from

Journal of Vision (2010) 10(4):16, 1–17 1

doi: 10 .1167 /10 .4 .16 ISSN 1534-7362 * ARVOReceived November 26, 2009; published April 28, 2010

http://journalofvision.org/10/4/16/

irrelevant low-level differences between the stimuli.However, other studies pointed out that a considerablepart of the differential activity occurring at short latencieswas not affected by the status of the stimulus (Johnson &Olshausen, 2003, 2005). This phenomenon was partic-ularly marked in a study using human and animal faces asstimuli. Subjects were extremely good at respondingselectively to either human or animal faces and couldswitch virtually effortlessly between the two differenttarget categories from block to block (Rousselet, Mace, &Fabre-Thorpe, 2003), achieving accuracy levels of over97% correct, irrespective of whether the target categorywas human or animal. Despite this, the analysis of thesimultaneously recorded ERP signals failed to find anyevidence that short latency responses (i.e., at latenciesbelow about 180 ms) could be modulated by whether theparticular stimulus was a target or not (Rousselet, Mace,Thorpe, & Fabre-Thorpe, 2007). There were strongdifferences in the ERP signals recorded with human andanimal faces, but those differences were unaffected bywhether the subject was treating the image as a target ornot. If a dependence on task status is needed to be able toinfer that a particular neural response is related to high-level processing, then it would be natural to conclude thatno influence of such high-level factors is visible untilaround 180 ms in this task.Does this mean that the differential brain responses seen

earlier should be dismissed as simple artifacts due toirrelevant low differences between images? One type ofresult that argues strongly against this are recent studiesusing a saccadic choice task that show that usefulbehavioral responses can be generated well before the150–180 ms latency value suggested by the task-dependentERP effects. In the first such study, Kirchner and Thorpereported that when two natural scenes are simultaneouslyflashed left and right of fixation, reliable saccades to imagescontaining animals can be initiated as early as 120–130 msafter images onset (Kirchner & Thorpe, 2006). Given thatmotor preparation presumably needs at least 20 ms, thisimplies that the underlying visual processing may needonly 100 ms, considerably earlier than the 150-ms latencyof the first differential activity.In the present study, we have used a similar saccadic

choice task protocol to the one used previously byKirchner and Thorpe, but with another type of highlysignificant visual stimulusVphotographs of human faces.Experiment 1 directly compared saccadic reaction timesfor three stimulus categoriesVanimals, faces, and vehi-clesVand demonstrated a very impressive result for facesfor which the fastest reliable saccades were seen atlatencies as early as 100–110 ms. Then, Experiment 2revealed a search asymmetry between faces and vehicles:subjects found it much easier to saccade toward the facesthan toward the vehicles. Interestingly, they were onlyable to fully overcome this bias for saccades withrelatively long latencies. Finally, Experiment 3 allowedus to test a simplistic model of the task based on a

comparison between activation in the left and righthemispheres. It showed that the fast detection of faces isnot restricted to a horizontal display arrangement andmust presumably rely on a more complex process than asimple comparison between activation levels in the twohemispheres.The results, some of which have been presented

previously (Crouzet, Kirchner, & Thorpe, 2008), fit witha large range of previous studies that have demonstrated thatfaces have a special status. More importantly, they imposesome very serious temporal constraints on the underlyingprocessing. Specifically, face-selective behavior can beinitiated at latencies so short that there may not be enoughtime to complete the first wave of processing in the corticalareas of the ventral stream. If so, other possibilitiesincluding the involvement of subcortical pathways may beinvolved.

Experiment 1: Comparisonbetween three target categories

Experiment 1 examined whether performance in theSaccadic Choice Task varies when different objectcategories are used as the target (faces, animals, orvehicles). Each of these target categories was displayedin combination with the same set of “neutral distractors”,corresponding to various natural scenes. Thus, any differ-ences observed between the three conditions will reflectdifference of processing time between the three categoriesof objects, rather than differences between the distractorstimuli.

MethodsParticipants

Eight volunteers (7 men; mean age 24.5 years, rangingfrom 22 to 31 years) with normal or corrected-to-normalvision participated in a 2-AFC saccadic choice task. Theyall gave written informed consent to participate in theexperiment.

Stimuli

One thousand photographs selected from the CorelPhotolibrary database or downloaded from the Internetwere used to set up four object categories of 250 naturalscenes: faces, animals, vehicles, and neutral distractors.The neutral distractor category was composed of a rangeof images that all contained a salient object in theforeground. All the images were converted to grayscaleand resized to 330 � 330 pixels. The global contrast ofeach image was reduced to 80% of the original image,

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 2

allowing us to adjust the mean luminance of each image toa grayscale value of 128. (Guyonneau, Kirchner, & Thorpe,2006). The complete set of images can be provided onrequest (Figure 1).

Apparatus

Participants viewed the stimuli in a dimly lit room withtheir heads on a chin rest to maintain the viewing distanceat 60 cm. Stimuli were presented on an IIYAMA VisionMaster PRO 454 monitor with the screen resolution set to800 � 600 pixels and a refresh rate of 100 Hz. The centersof the two images were always 8.6- from the fixationcross, resulting in a retinal size for each image of 14- by14-. The experiment was run using the software Presenta-tion 9.9 (Neurobehavioral Systems).

Protocol

The experiment was performed using a Saccadic ChoiceTask, a similar protocol to the one used by our team inprevious studies (Guyonneau et al., 2006; Kirchner &Thorpe, 2006) with the exception that the natural sceneswere displayed during 400 ms rather than flashed for20 ms. The original reason for using such short presentation

times when using a conventional manual go/no-go proto-col was to exclude the possibility of ocular exploration(Thorpe et al., 1996). However, in the present experi-ments, we specifically required the subjects to makesaccades. Since we were recording the eye movements,the original argument for using flashed presentations wasnow obsolete. Preliminary experiments had already shownthat, contrary to what might have been expected, thislonger presentation durations significantly shortened themean RT. Specifically, it appears that using longerpresentation times reduces the number of late saccades,resulting in a leftward shift and a sharpening of theSaccadic Reaction Time (SRT) distribution. An explan-ation could be that the offset of the images in the originalprotocol perturbed the initiation of some saccades,resulting in a right-biased distribution. A second differ-ence with respect to the previous studies was that thebackground screen was set at a grayscale value of 128rather than black.Observers had to keep their eyes on a black fixation

cross that disappeared after a pseudo-random time interval(800–1600 ms), leaving a 200-ms time gap before thepresentation of the images (Fischer & Weber, 1993;Kirchner & Thorpe, 2006). The use of such a gap allowssaccades to be initiated more rapidly. Two natural scenes,

Figure 1. Examples of images used in this study.

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 3

one target and one distractor, were then displayed on eachside of the screen for 400 ms (see Figure 2). The task wasto make a saccade as quickly and as accurately as possibleto the side where an object belonging to the targetcategory has appeared.Using a within-subject design, three object categories

were tested here (faces, animals, and vehicles). Eachsubject saw each target image once during the experimentand thus performed 250 trials in each condition, dividedinto blocks of 50 trials. The “neutral distractor” imageswere the same in the three conditions, so each one wasseen several times by each participant. The order of thethree conditions was counterbalanced across participants.Each block was preceded by a training session of 50 trialswith images not used in the experiment.

Response recording and detection

Eye position was recorded using horizontal EOGelectrodes (1 kHz, low pass at 90 Hz, notch at 50 Hz,baseline correction [j400:0] ms; NuAmps, Neuroscan).Saccadic Reaction Time (SRT) was determined offline asthe time difference between the onset of the images andthe start of the saccade. Each trial was verified by theexperimenter to make sure that only the largest inflection(if any) was taken as a real saccade (see Kirchner &Thorpe, 2006 for more detailed information about theprocedure); 15.2% of trials (912 of 6,000) had to be

excluded because of a noisy eye signal, but this percentagewas evenly spread across conditions (face task = 16%,animal task = 14.6%, vehicle task = 15%).

Minimum reaction times

To determine a value for the minimum SRT, we dividedthe saccade latency distribution of each condition into10-ms time bins (e.g., the 120-ms bin contained latenciesfrom 115 to 124 ms) and searched for bins containing sig-nificantly more correct than erroneous responses using a#2 test with a criterion of p G 0.05. If 5 consecutive binsreached this criterion, the first was considered to corre-spond to the minimum reaction time.

Results and discussion

The principal finding from this first study was thatsubjects were fast and accurate in all three conditions (see

Figure 2. Protocol: The saccadic choice task. Observers had tofixate a cross in the center during a pseudo-random time (800–1600 ms). After a gap of 200 ms, 2 images were displayed left andright of fixation for 400 ms. Observers then had 1000 ms toprepare for the next trial.

Figure 3. Experiment 1. (Top) Distributions of SRT for 3 differenttarget categories: face, animal, vehicle. Correct responses areshown in thick lines, incorrect as thin lines. (Bottom) Mean accuracyand SRT in the 3 conditions. Errors bars are SEM.

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 4

Figure 3). Ultra-rapid processing of objects in the saccadicchoice task is clearly not restricted to animals but can alsobe extended to vehicles and human faces. However, andcontrary to what has been shown using manual response(Rousselet et al., 2003; VanRullen & Thorpe, 2001a), ourresults showed a clear ordering between categories forboth SRT and accuracy. A one-factor ANOVA analysisshowed that there was a global effect of the category usedas target on both mean SRT (F(2,14) = 10.622, p G 0.01)and accuracy (F(2,14) = 26.031, p G 0.001). A post-hocTukey analysis for multiple comparisons showed aprogressive increase on accuracy from the “vehicle”condition (75%) to the “animal” (82.4%) and “face”(94.5%) conditions. Mean SRT was only significantlydifferent between the face (147 ms) and vehicle (188 ms)conditions.

Different processing times for different objectcategories

In addition to the very clear differences in meanreaction times seen for the three object categories, therewere also very striking differences in the minimumreaction time values (i.e., the first bin of at least 5consecutive bins in the reaction time distribution wherethere was a significantly higher proportion of correct thanerroneous responses). In the case where the targetcategory was “animal”, the value obtained replicatedKirchner and Thorpe’s (2006) study with a minimumSRT of 120 ms. Interestingly, this minimum SRT wasclearly higher for vehicles (140 ms) and lower for faces(110 ms). Together, the results demonstrate a very clearadvantage for the processing of faces over animals andvehicles when observers had to discriminate these objectcategories from “neutral distractors”.

Experiment 2: Faces vs. vehicles

Previous studies using a manual go/no-go protocol hadshown that subjects can switch from one target category toanother in different blocks with little cost in terms ofeither accuracy or reaction time. This was seen for boththe situation where the target categories are animals andmeans of transport (VanRullen & Thorpe, 2001a), as wellas with humans versus animals (Rousselet et al., 2003). InExperiment 2, we ask whether this ability to switchbetween target categories also exists in the case of thesaccadic choice task. Specifically, we designed an experi-ment in which the subjects had to discriminate directlybetween two object categories: faces and vehicles. Thisdesign allowed us to directly compare processing timesbetween these two categories of objects, but additionally,we can see if the task can be reversed under voluntarycontrol. As an example, if a subject was instructed to treat

faces as targets and vehicles as distractors in a first block,subsequent blocks could require the reverse configuration,with vehicles as targets and faces as distractors. Theresults reveal a clear asymmetry, with saccading to facesbeing considerably faster and more accurate than tovehicles. Indeed, subjects had great difficulty in makingfast saccades toward vehicle targets.

MethodsParticipants

Eight volunteers (5 men; mean age 26.9 years, rangingfrom 23 to 34 years) with normal or corrected-to-normalvision participated in a 2-AFC saccadic choice task. Theyall gave written informed consent to participate in theexperiment.

Stimuli

In order to have a more controlled set of stimuli, 200photographs were selected from the Corel database andthe Internet to generate two object categories, each with100 images: faces and vehicles. The faces were 50% menand 50% women, while the vehicles were 50% cars and50% trains. Each subcategory was divided equally intoclose-up and mid-distance views. Manipulations on theluminance and contrast of each image were the same as inExperiment 1.

Apparatus and protocol

The design was unchanged from Experiment 1 with thefollowing exception: each subject saw each image fourtimes, both as target and distractor and in both the left andright hemifields. Specifically, each participant performed200 trials in each of the two conditions (face and vehicle).The order of the two conditions was counterbalancedacross participants; 23% of trials (746 of 3,200) had to beexcluded because of a noisy eye signal, but this per-centage was evenly spread across conditions (face task =22.3%, vehicle task = 24.4%).

Results and discussion

The first main result of Experiment 2 is that even withanother object category as distractor, here vehicles,saccades toward faces can still be initiated very rapidly.Overall accuracy was 89.6%, with a mean SRT of 138 ms.Remarkably, the earliest reliable saccades appeared just100–110 ms after scene onset. However, the situation wasmarkedly different when the target category was vehicle.In this case, the values for mean SRT (167 ms) andaccuracy (71%) were both significantly poorer than withfaces (F(1,7) = 84.723, p G 0.001 and F(1,7) = 44.867, p G0.001, respectively; Figure 4).

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 5

Even more striking was the distribution of response inthe 100–140 ms time window. There was a tendency ofsaccades initiated in this time range to go toward the sidewith the faces, even if the task is to go to vehicles. Thus,saccades initiated before 140 ms seemed to be hard tocontrol.An interesting observation is that if we divide the data

according to the position of the target (left or right,Figure 5), there is a clear tendency for subjects to be fasterand more accurate when the target is on the left. Thistendency can be observed in the face task (left: 135 msand 95.8%; right: 143 ms and 83.3%) and in the vehicletask (left: 165 ms and 77.2%; right: 170 ms and 64.8%)and is significant for mean RT (F(1,7) = 6.4953, p G 0.05)although not for accuracy (F(1,7) = 4.0321, p = 0.084).Thus, if the target is in the left hemifield, participants

produced fewer errors and their correct responses had ashorter mean SRT. When the target was in the righthemifield, participants made more early errors. Further-more, many of the errors in the face task were made whenthe target is on the right and the distractor on the left,especially on fast saccades. For example, the specificpattern observed in Figure 4 with more responses towardfaces than vehicles in the 100–140 ms time window islargely the result of the situation when the face is on theleft. A similar tendency to produce more saccades on theleft has also been reported when people look at chimericfaces (Butler et al., 2005). These left hemifield biasescould be related to the well-known fact that neuralresponses in the right hemisphere are reliably strongerthan on the left, a result that has been repeatedly seen inboth fMRI studies (e.g., Hemond, Kanwisher, & Op deBeeck, 2007; Kanwisher, 2000) and ERPs (e.g., Jacques &Rossion, 2009; Rousselet, Mace, & Fabre-Thorpe, 2004).Indeed, the existence of a left hemifield advantage whensaccading to faces supports the hypothesis that the saccadicchoice task really does involve face processing mecha-nisms. If it were simply a bias toward making saccadestoward the left, the same biases would be expected with anysort of target.Additionally, half the stimuli were close-up views (CV),

and the other half mid-distance views (MV). This alloweda post-hoc analysis of the effect of object size to beperformed. It showed that there was absolutely no effectof the size of the target face on either SRT or on accuracy.This effect is null for the “face” (CV: 168 ms and 70%;MV: 165 ms and 72%) as well as for the “vehicle”conditions (CV: 138 ms and 90%; MV: 137 ms and 89%).Further studies to look at the effect of object size ondiscrimination performance in this task in a more system-atic way would be of considerable interest.In summary, performance in this task was remarkably

fast and efficient. Subjects were able to move their eyesselectively to the side where the scene contains a facecategory target in as little as 100–110 ms. It is clearthat we would not argue from this result that a face canbe “recognized” in just 100 ms. Nevertheless, the resultdemonstrates that the visual system needs only around100 ms to initiate an eye movement toward a face.Furthermore, the fact that subjects had such difficulty inreversing the task and saccading toward the vehicle sug-gests that this attractivity might be effectively hard-wired.

Experiment 3: Horizontal versusvertical positioning

One factor that might contribute to this remarkable levelof performance may lie in the design of the protocol, inwhich the two images are displayed to the left and right offixation. As a result, the two images will effectively be

Figure 4. (Top) Distribution of SRT over all subjects when the taskis to saccade toward faces (responses toward faces in orange,vehicles in blue). (Bottom) Distribution of SRT for all subjectswhen the task is to saccade toward vehicles. The gray vertical barindicates the bin where correct responses start to significantlyoutnumber errors.

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 6

processed separately by the two hemispheres (at leastinitially) and this may provide a situation that isparticularly favorable. Potentially, the task could beperformed by comparing the activation in the two halvesof the brain, and initiating the saccade to the side that hasthe strongest (or earliest) activation. This hypothesis wastested in Experiment 3 in which subjects were asked toperform the same task with either the images displayedhorizontally (to the left and right of fixation) or vertically(above and below the fixation point). Experiment 3 alsoexamined the difference between a saccadic choice task inwhich two images are presented at the same time andsubjects required to saccade to a target, and simpledetection task in which only one image is presented oneach trial. As in Experiment 2, subjects were required toperform the task with faces and vehicles images, varyingthe target category in different blocks.

MethodsParticipants

Four volunteers (2 men; mean age 32.7 years, rangingfrom 23 to 50 years) with normal or corrected-to-normal

vision participated in a 2-AFC saccadic choice task. Theyall gave written informed consent to participate in theexperiment.

Stimuli

Unchanged from Experiment 2 (Figure 6).

Protocol

The protocol was unchanged from Experiments 1 and 2with the following exceptions. The experiment wasdivided in two sessions, each comprising 12 blocks of50 trials. The first two blocks and the last two blocks ineach session used a Simple Detection Task with only oneimage on each trial and no distractor, where the twocategories of target (faces and vehicles) were mixed in thesame block. Within each two-block group, one block usedthe images in the horizontal arrangement, while the otherused a vertical one. For this Simple Detection Task,subjects were instructed to saccade as fast as possible onthe side where there was an image, independently of thecategory of the image. It is important to notice that in thecase of this Detection Task, no classification was needed.

Figure 5. Distributions of SRT over all subjects when the task is to saccade toward faces (top row) or vehicles (bottom row) and when thetarget is on the left (left column) or on the right (right column). Correct responses are in thick lines, incorrect are in thin lines.

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 7

In the middle part of each session, the subjects performedthe Saccadic Choice Task with either four blocks withFaces as targets followed by four blocks with Vehicles astargets, or the contrary. In addition, the arrangement of thestimuli was varied with two blocks of vertically arrangedstimuli alternating with two blocks using the horizontalarrangement. All the different orders were counterbalancedacross the four subjects and the two sessions. Image size

was still set at 330 � 330 but the resolution of the screenwas increased to 1024 � 768 to allow the images to bedisplayed vertically. As a consequence, the retinal size ofthe images was 11- by 11-, and the center of images was6.8- from the center of the screen.

Eye movement recording

Unlike Experiments 1 and 2, in this experiment the eyemovements were monitored using a Chronos Eye Tracker(Chronos Vision, Berlin, Germany). This infrared trackingsystem samples eye position at 200 Hz binocularly.Saccade detection was performed offline, based on avelocity criterion and all the saccades were verified by theexperimenter. Only the first saccades to end beyond 4- ofeccentricity (corresponding to a minimum of 25% of thewidth of the image) were included in the analysis; 10.9%of trials had to be excluded using this criterion or becauseof a poor detection of the pupil. Before each block, an8-point calibration procedure was performed.

Results and discussionVertical vs. horizontal display

As can be seen from Table 1, performance when theimages were arranged vertically remained very good andwas quite similar to the results obtained with the horizontalarrangement. Indeed, there was no overall effect of thearrangement of the stimuli (horizontal or vertical) onaccuracy (as demonstrated by a two-way ANOVA). Incontrast, the mean RTs for horizontal saccades were sig-nificantly shorter than vertical ones (167 ms and 178 ms,respectively; F(1,9) = 9.1069, p G 0.05). As might beexpected from the results of Experiments 1 and 2, thenature of the target (face or vehicle) still has a strong

Figure 6. Design of Experiment 3. The protocol was similar to theone used in Experiments 1 and 2. After the 200-ms gap, andfollowing a block design, participants had to perform a task inwhich either one image was presented (two screens on thebottom of the figureVSimple Saccadic Detection Task) or twoimages are presented simultaneously (two screens on the top-Saccadic Choice Task). In both cases, the images can bedisplayed horizontally or vertically.

Target category Target location

Saccadic choice task Simple detection task

Mean SRT (ms) Accuracy (%) Min. SRT (ms) Mean SRT (ms) Min. SRT (ms)

Face Left 150 T 14 95.5 T 2 119 T 12Right 159 T 13 84.3 T 5.9 133 T 14Horizontal display 154 T 13 89.8 T 3 100 126 T 12 80Bottom 165 T 17 85.4 T 6.5 146 T 11Top 168 T 11 86.6 T 5.8 135 T 10Vertical display 166 T 14 86.3 T 4.2 110 139 T 10 90

Vehicle Left 176 T 17 83 T 9 125 T 15Right 185 T 19 68.8 T 9.1 147 T 16Horizontal display 180 T 17 75.8 T 6.8 170 136 T 15 80Bottom 190 T 20 67.1 T 8 140 T 13Top 187 T 13 74.3 T 5.8 138 T 10Vertical display 189 T 16 71 T 4.7 190 139 T 11 90

Table 1. Results for Experiment 3. Mean SRT and accuracy are presented for both the Saccadic Choice Task (with two simultaneouslypresented images) and the Simple Detection Task (a single image presented).

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 8

effect both on mean RT and accuracy (F(1,9) = 48.7835,p G 0.001; F(1,9) = 31.5336, p G 0.001, respectively).Thus, these results showed a slight difference between ahorizontal and a vertical display.However, a more detailed analysis that divided the

results according to the precise location of the target (Left,Right, Top, Bottom) revealed that the difference betweenthe results with horizontal and vertical displays wasessentially due to the strong advantage when the target ispresented on the left, an effect already seen in Experiment 2.Thus, it seems that in these experiments, saccading to theright, the top, or the bottom of the screen was roughlyequivalent, but that saccading to the left was faster.The critical point here was that subjects were still able

to produce very fast responses in the saccadic choice taskeven when the stimuli were positioned vertically. Thiseffectively rules out a simple “hemisphere comparison”hypothesis in which the eyes simply move because there isan imbalance of activity between the left and righthemispheres. This is because, presumably, when theimages are positioned vertically the amount of activationin the two hemispheres will be roughly balanced. How-ever, it is worth noting that there is also evidence that theprocessing of the upper and lower visual fields involvesanatomically separate areas in extrastriate cortex. As aconsequence, it may still be possible to envisage acompetitive mechanism in which global activation levelsin two separate brain structures are compared. Furtherexperiments would be needed to test the limits of this sortof ability by, for example, presenting both the target anddistractor stimuli in the same hemifield, or even in thesame quadrant.

Saccadic choice task vs. simple saccadic detectiontask

A second major result emerging from Experiment 3 wasthe relatively small difference between SRT distributionsin the simple saccadic detection task and the saccadicchoice task, at least with faces as targets. The fact thatthere are essentially no errors in the simple saccadicdetection task means that it is not useful to compareaccuracy levels. Mean SRT values were 159 ms for facesversus 183 ms for vehicles in the saccadic choice task.The corresponding values were 131 ms and 137 ms in thesimple saccadic detection task. A two-way ANOVAshowed that there is still a significant effect of thecategory of the target (F(1,9) = 22.9889, p G 0.001), anda clear advantage for simple detection over the saccadicchoice task (F(1,9) = 144.6105, p G 0.001). Furthermore,the interaction between the task and the category of thetarget is significant (F(1,9) = 9.0315, p G 0.05), meaningthat the difference between saccadic choice and simpledetection is much larger for vehicles than for face targets.This is also clear from looking at the minimum SRTbecause when the target was a face, the difference between

minimum SRTs for the choice task and the simple detectiontask was only 20 ms. In comparison, when the target wasa vehicle, this additional time cost was 80 ms.

General discussion

The present study used a saccadic choice task toinvestigate the time course of the processing involved inultra-rapid detection of objects in natural scenes. Theexperiments follow on from an earlier study that hadshown that subjects can make rapid and reliable saccadestoward animal targets when two images are simultane-ously flashed left and right of fixation (Kirchner &Thorpe, 2006). Experiment 1 showed that these very rapidsaccadic responses can be initiated even more rapidlywhen human faces are the target category, with the fastestsaccades being initiated from around 110 ms following theonset of the stimuli. Then, Experiment 2 showed that thisstrong bias toward saccading toward faces is very difficultto suppress, because even when subjects are activelytrying to saccade toward vehicles, they still show a veryclear tendency for fast saccades to be directed towardfaces. Finally, Experiment 3 showed that this ability toinitiate very fast saccades toward face targets is notrestricted to the specific left/right design and thus cannotbe explained by a simple comparison between activationlevels in the two hemispheres.

Ultra-rapid processing of faces

The values reported here for saccadic reaction timeswith faces are remarkably short. In Experiment 2, inwhich faces were paired with photographs of cars andtrains, the mean onset latency for saccades toward faceswas a mere 138 ms. Virtually all the saccades wereinitiated in under 200 ms, but despite this, accuracy was avery respectable 89.6%. By examining the distribution ofcorrect and erroneous saccades in each 10-ms bin of thereaction time distribution, we were able to show that theminimum reaction time in this task was only 100–110 ms.Such values put very severe temporal constraints on theunderlying visual processing, especially when one takesinto account the fact that the latencies obtained from theEOG and eye tracker data used in this study include thetime needed to initiate the eye movement. Most neuro-physiologists would allow roughly 20 ms for the activa-tion of the brain stem structures involved in oculomotorcontrol and the muscles of the eye itself. If true, then itappears that information about the presence of a face in animage may be available as little as 80 ms after the onset ofthe image. Such values are considerably shorter than

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 9

previous behavioral estimates of processing times in thehuman visual system (Thorpe et al., 1996) and are even alot shorter than the 150-ms differential ERP responsebetween targets and distractors that has previously beenused as a measure of processing time.The mean SRT values seen here are substantially

shorter than those reported in the previous study byKirchner and Thorpe (2006) in which the median SRTwhen using photographs of animals as targets was 228 ms.However, there are a number of differences between thetwo experiments that could explain why the reaction timeswere so much shorter here. In Experiment 1, we used avery similar situation with photographs of animals pairedwith complex natural scenes as distractors and obtained amean SRT value of 170 ms. It seems likely that some ofthis reduction in reaction time results from the fact that inthe current experiments, the two images remain presentfor 400 ms, instead of simply being flashed for 20 ms, asin the original study. It could be that removing thestimulus before the saccade has been initiated in theoriginal design might interfere with saccade initiation. Incontrast, by leaving the images on for 400 ms, the subjectis in the very natural situation of initiating saccadestoward a stimulus that is still present when the eyes arriveat their destination. Given that the aim of the experimentis to obtain the shortest realistic measurement of the timerequired to initiate a saccade toward a target, there seemsto be little reason to continue with the original design,which only seems to introduce additional variability in thereaction time distribution.

Comparison with previous eye movementstudies

The tendency of humans to look preferentially at faceswhen exploring visual scenes was already clear in theclassic studies of Buswell (1935) and Yarbus (1967), anda recent study showed that similar biases also exist inchimpanzees (Kano & Tomonaga, 2009). However, it isimportant to realize that several factors may be involvedin producing such biases (Henderson, 2003). For example,there may be a tendency to fixate faces for longer thanother less interesting parts of the image. The most wellwidely used models of gaze control, such as Itti andKoch’s (2000) saliency model rely on local variations inrelatively low-level factors such as color, orientation, andluminance. While such models can account for a sub-stantial proportion of real-world gaze patterns in humans(Parkhurst, Law, & Niebur, 2002; Peters, Iyer, Itti, &Koch, 2005; Tatler, Baddeley, & Gilchrist, 2005), thereare a number of studies showing that such models cannot beconsidered complete (Birmingham, Bischof, & Kingstone,2008; Cerf, Harel, Einhauser, & Koch, 2008). Indeed, bychanging the task requirements, it is possible to overridethese low-level biases (Einhauser, Rutishauser, & Koch,2008).

A separate question concerns the issue of whether thevery first saccades that are generated in response to ascene can be directed to important objects such as faces,and if they can, at what latency. The study by Kirchnerand Thorpe had already demonstrated this for animaltargets, and a recent study showed that these rapidsaccades toward animals can be quite accurate in terms oflocalization (Drewes, Trommershaeuser, & Gegenfurtner,2009). Another recent study by Fletcher-Watson, Findlay,Leekam, and Benson (2008) extended the use of the sac-cadic choice task to the detection of humans. Observersviewed two images presented at the same time on the leftand right of a screen, only one of which contained ahuman. They reported that participants tend to saccademore on the side with the human, and that this bias wasindeed seen for the very first saccades. The saccades weregrouped into bins of 50 ms, and not surprisingly, the firsttwo bins had very few responses. However, in the binfrom 100 to 149 ms, 90% of the saccades were orientedtoward the image containing a face. One particular featureof that study was that they included a task where thesubjects had to saccade spontaneously to one of twoimages, with no task to perform. The fact that even heresubjects showed a strong tendency to look toward the sidewith the human suggests that there may well be a built-inbias toward looking at humans.Other data supporting the idea that faces can be

processed very efficiently comes from the extensiveliterature on attentional capture (Bindemann, Burton,Hooge, Jenkins, & de Haan, 2005; Langton, Law, Burton,& Schweinberger, 2008; Ro, Russell, & Lavie, 2001;Theeuwes & Van der Stigchel, 2006; Vuilleumier, 2000)as well as recent studies reporting pop-out in displayscontaining large numbers of elements (Hershler &Hochstein, 2005, 2006; although see also Brown, Huey,& Findlay, 1997; Vanrullen, 2006). Additionally, the biastoward faces has also been seen using an anti-saccadeprotocol, which demonstrated that subjects have difficultyin looking away when a face is present (Gilchrist &Proske, 2006). Together, all these results point toward areal behavioral advantage for faces in a wide range ofsituations.

Underlying brain mechanisms

These behavioral effects are also reflected at the level ofbrain mechanisms. Numerous studies have suggested thatfaces may have a special computational status that wouldallow them to be processed more efficiently and fasterthan other classes of objects (Farah, Wilson, Drain, &Tanaka, 1998; Haxby, Hoffman, & Gobbini, 2000;Kanwisher, 2000; but see Tarr & Gauthier, 2000). Facesare known to activate spatially adjacent but distinct brainregions in both humans and monkeys (Freiwald, Tsao, &Livingstone, 2009; Sergent, Ohta, & MacDonald, 1992;Tsao & Livingstone, 2008). This has been shown in a

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 10

range of techniques including PET, fMRI, and single unitrecording (Freiwald et al., 2009; Ishai, Ungerleider, Martin,Schouten, & Haxby, 1999; Kanwisher, McDermott, &Chun, 1997; Puce, Allison, Gore, & McCarthy, 1995).Nevertheless, there is still debate about the nature anddegree of specialization (Cohen & Tong, 2001; Downing,Jiang, Shuman, & Kanwisher, 2001; Haxby et al., 2001).The fact that the earliest reliable saccades toward faces

can be seen as early as 100–110 ms after stimulus onsetplaces particularly severe constraints on the underlyingbrain mechanisms. In the present context, it is particularlyimportant to look at experimental evidence on the speedwith which information about faces can be processed.Following the earliest reports of face-selective EventRelated Potentials (Jeffreys, 1989), much attention has beenpaid to the N170 potential that seems to be particularlystrongly associated with face processing (Bentin, Allison,Puce, Perez, & Mccarthy, 1996; McCarthy, Puce, Belger,& Allison, 1999; for a recent review, see Rossion &Jacques, 2008). However, it seems likely that the N170occurs too late to be directly involved in triggering thefastest saccades reported here. Nevertheless, there havebeen repeated reports of face-selective electrophysiologicalresponses occurring at even earlier latencies. For example,Liu, Harris, and Kanwisher (2002) reported face-selectiveMEG activation at latencies of around 100 ms, and selec-tive ERP responses to emotional faces have been reportedwith latencies of around 120 ms (Eimer & Holmes, 2002).There have even been reports of face-selective repetition-related effects at even shorter latencies, sometimes asearly as 45–80 ms (George, Jemel, Fiori, & Renault, 1997;Mouchetant-Rostaing&Giard, 2003; Mouchetant-Rostaing,Giard, Bentin, Aguera, & Pernier, 2000) or even 30–60 ms(Braeutigam, Bailey, & Swithenby, 2001), but it has beenunclear whether these very rapid differential effects arereally related to face perception. Clearly, in the light of thepresent behavioral responses, it may be appropriate toreconsider the significance of these very early phenomena.Another important source of information about process-

ing speed is the results of single-cell recording studies inawake primates that have shown that face-selectiveneuronal responses can be seen from around 100 msalthough the fastest single unit responses can be as earlyas 70 ms (Oram & Perrett, 1992). It is also important todetermine precisely when information about object iden-tity can be read out from the activity of a population ofcells. This issue was addressed in a recent study thatexamined the responses of populations of single neuronsin monkey inferotemporal cortex and found that decisionsabout both object category and identity could be made onsingle trials from around 100 ms using a temporal windowof only 12.5 ms in duration (Hung, Kreiman, Poggio, &Dicarlo, 2005).In contrast to work in monkeys, there have been

relatively few single unit recording studies in humans(though see, for example, the work that has been done onrecordings from the medial temporal lobe in epileptic

patients; Mormann et al., 2008). Most human data comesfrom ERP and MEG recordings that are less easy to relatedirectly to behavioral reaction times because their analysisrequires pooling together responses from a very largenumber of trials. The problem is that while a significanteffect may be detected in the pooled data at a givenlatency, this does not mean that enough information couldbe extracted in real time on a single trial as would berequired to initiate a behavioral response. However, arecent study of local field potential recordings obtainedfrom occipito-temporal cortex in human epileptic patientsreported that even in humans, information about objectcategory can reliably be derived on a single trial basisfrom only 100 ms after stimulus onset (Liu, Agam,Madsen, & Kreiman, 2009). The study also showed thatface-selective intracerebral responses were remarkablyinvariant to changes in size, position, and viewpoint, evenat such short latencies. While these neurophysiologicalstudies provide clear evidence that information about thepresence of (for example) a face can potentially beextracted from brain activity shortly after stimulus onset,the current behavioral data goes further by demonstratingthat such information can indeed be used by the brain tocontrol behavior. Furthermore, while information can beextracted from intracortical potentials in humans from100 ms (Liu et al., 2009), this does not imply that theinformation is necessarily already visible in neuronal firing.For example, in monkey IT, the earliest face-selectivefiring has been reported at latencies of 70–90 ms (Oram &Perrett, 1992), but deflections in intracerebrally recordedpotentials are seen from as early as 50 ms (Schroeder,Mehta, & Givre, 1998). Thus, it might be that face-selective firing in the brain regions studied by Liu et al.might not occur until appreciably later, perhaps 120–30 ms, which would be well after the earliest face-selectivesaccades reported here.While it may seem natural to assume that the processing

required to initiate these fast responses involves theventral cortical processing stream, it is important torealize that there is little reason to exclude subcorticalprocessing pathways. For example, there is certainlyevidence that face information can be processed insubcortical structures such as the amygdala, and there isevidence that visual information can reach the amygdalavia the superior colliculus and the pulvinar (Johnson,2005). Much of the evidence for subcortical processinghas come from work with emotional faces and fear-inducing stimuli (Ohman, Carlsson, Lundqvist, & Ingvar,2007), but it seems clear that the fast saccadic responsesthat we describe here are not restricted to faces withemotional expressions. Nevertheless, there is no strongargument for excluding the possibility that at least someof the information needed for face detection might have asubcortical origin.One further result from the monkey neurophysiology

that seems particularly important in the present context isthe finding that the onset latencies of neurons in

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 11

inferotemporal cortex can vary significantly depending onthe stimulus. Kiani, Esteky, and Tanaka (2005) reportedthat onset latencies of responses to primate and humanfaces are roughly 20 ms earlier than to faces of otheranimals. This difference was seen even though the totalamount of activity is similar for the two types of stimulus,demonstrating that it is not simply that the neuronsrespond better to primate and human faces.This result suggests an interesting possibility that could

go some way toward explaining the remarkably rapidsaccadic responses reported here. Suppose that when aface and a vehicle are simultaneously presented in the leftand right visual fields, the neurons in the ventral streamcontralateral to the face fire 20 ms earlier than the neuronsresponding to the vehicle. This may produce an imbalancein the levels of activation, which could result in differ-ential activation in areas involved in saccade initiationsuch as the Frontal Eye Field (FEF) and the LateralIntraparietal Area (LIP). It is as if the face would have ahigher salience than other stimuli, simply because of thisdifference in onset latency.If there is indeed a difference in the onset latency of

neuronal responses to different types of stimuli, with theshortest latencies being seen for faces, then this mightwell explain why such short latency behavioral responsescan be seen to faces. However, such an explanation wouldlead to the following hypothesis. Suppose that it is notpossible to alter the latency of the neural responses undertop-down control. In this case, we might expect that evenif the subject was trying to direct their eyes toward the otherstimulus, the latency advantage for faces would still resultin a bias toward faces, at least for the fastest responses. Thisis precisely what we observed in Experiment 2, where wenoted that when the subjects were instructed to saccadetoward the vehicle, they nevertheless tend to saccadetoward the face, at least when the saccades were initiatedin under 140 ms.

A difference between manual and saccadictasks?

The strong bias toward responding to faces reportedhere is not something that we have seen in previousstudies using manual responses. These earlier studiesusing similar visual stimuli reported that subjects couldeasily shift between one target category and another fromblock to block. This was seen for both animals and meansof transport (VanRullen & Thorpe, 2001a) and for animaland human faces (Rousselet et al., 2003). However, inthese studies using a manual go/no-go task, even the veryfastest responses are never seen earlier than about 250–300 ms following stimulus onset, considerably longer thanthe saccadic responses reported here. This raises thepossibility that the extra processing time available in themanual task allows the subjects to generate a behavioralresponse that is fully under top-down control. In contrast,

eye movements may be generated without allowing enoughtime for complete modulation of the behavior. This is clearfrom the results of Experiment 2 in which participants triedto selectively make saccades toward the vehicle. Thispattern of results is precisely what would be expected ifthe visual system had an in-built bias in favor of faces,which would be evident even under conditions where thetask requires the participants to saccade elsewhere.The latency at which the saccades start to be modulated

by the task requirements could be related to neuro-physiological studies of attentional effects on neuronalresponses. In studies of visual responses, it has repeatedlybeen noted that the initial transient part of the responsetends to be relatively fixed and that task-related modu-lations only start to be clear after a further delay(Roelfsema et al., 2007; Treue, 2001). It is thereforepossible that the difference in latency between the earliestsaccades to faces (100–110 ms) and the earliest point atwhich subjects can start to reliably make saccades towardthe vehicle target (around 140–150 ms) could reflect theduration of this initial period of the neural response, whichappears to be relatively insensitive to top-down taskmodulation.Once this top-down modulation has started, a decision

mechanism that depended on the amount of cumulatedactivity originating from the two parts of the visual fieldwould progressively become more and more stronglybiased toward the intended target. As a consequence,saccades that are initiated at relatively long latenciescould be reliably made in the direction of the target, evenwhen the target is a vehicle. In the case of behavioralresponses that have even longer latencies, such as amanual go/no-go response, the accumulation of activitywill be sufficient to ensure that the response can be madeto whatever target category is currently in use.The fact that the effectiveness of top-down task-

dependent modulation is not fixed, but varies dependingon the latency at which the saccades are initiated, meansthat the saccadic choice task can be used to track the timecourse of top-down effects, something that may bedifficult or even impossible using conventional manualreaction time methodologies. By the time a manualresponse is initiated, the brain has had plenty of time tocomplete a wide range of operations, including attentionalmodulation. In contrast, the very fastest saccadicresponses appear to occur before these modulatory effectshad time to go to completion. Until now, the only methodsthat have been able to investigate this early time window(100–150 ms) have been techniques such as EEG, MEG,and single-cell recording, but the behavioral significanceof such effects have been difficult to establish.

What makes faces special?

It appears that faces may be in a class of their own intheir ability to trigger very fast saccades, and a natural

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 12

question concerns the origin of this advantage. Onepossibility is that faces are special because we have agreat deal of expertise in processing faces from an earlyage. Indeed, faces occupy a very important part of thevisual environment of the newborn child (Sinha, Balas, &Ostrovsky, 2007), and this increased exposure could leadto the development of selective mechanisms usingunsupervised learning (see, for example, Masquelier &Thorpe, 2007). On the other hand, there is considerableevidence that humans have an innately specified bias infavor of face stimuli (Johnson, Dziurawiec, Ellis, &Morton, 1991; Johnson & Mareschal, 2001), and recentwork suggests that this preference carries across tophotographs of chimpanzees with which we have had farless experience (Taubert, 2009). Another recent studyshowed that in a change-blindness paradigm, we are muchmore likely to notice a change involving an animal thanone involving a vehicle, even when the stimuli have beenmatched for size within the image (New, Cosmides, &Tooby, 2007). The authors interpreted their results asfavoring an ancestral priority for animals, includinghumans. Irrespective of the origin of this face bias, it willundoubtedly have a major impact on the way in whichhumans explore the visual environment.

Selectivity mechanisms

The final point concerns the mechanisms that can allowface-selective responses to be produced so rapidly. Ourability to initiate directed saccades toward faces as earlyas 100–110 ms after stimulus onset clearly leaves littletime for anything other than a feed-forward pass. Thispoint is strengthened by the fact that reaction times in thechoice task are only marginally longer than in a simpledetection task (see Experiment 3), meaning that theprocessing overhead associated with detecting the pres-ence of a face in an image can be no more than a few tensof milliseconds. It has been known for some time that asingle wave of spikes can be sufficient not only for facedetection (VanRullen, Gautrais, Delorme, & Thorpe,1998) but also for face identification (Delorme & Thorpe,2001). There is further recent evidence that purely feed-forward hierarchical processing mechanisms may besufficient to account for at least some forms of rapidcategorization (Serre et al., 2007). In the field of computervision, considerable progress has been made in developingalgorithms for detecting and localizing faces in naturalimages (Hjelmas & Low, 2001; Viola & Jones, 2004),some of which have found their way into consumerproducts such as digital cameras.Indeed, the shortest saccadic reaction times that we

report here appear to be so fast that there may even not beenough time to complete the first feed-forward passthrough the ventral processing stream. Remember thatsince we need to allow around 20 ms for response initiation,it seems inevitable that face-selective mechanisms must

become activated from as early as 80 ms followingstimulus onset. Even in monkeys, this would correspondto the earliest responses in inferotemporal neurons, but itneeds to be remembered that monkeys are known to besubstantially faster than humans on virtually all behavioraltasks, probably because their smaller heads lead to areduction in conduction delays.Given these constraints, it seems likely that the visual

system will effectively try to make use of any availablecues to the presence of a face in the image, especiallythose that can be extracted early on during visualprocessing. Some recent research suggests that this mayindeed be the case. For example, Dakin and Watt (2009)have shown how the horizontal orientation structure in thehuman face provides a form of “bar code” that can beused for judgments of face identity. Further evidence forthe use of very low level heuristics comes from anotherstudy from our laboratory, that also made use of thesaccadic choice task, two images with varying contrastwere presented on the left and right, and participants wererequired to saccade to the one with the highest contrast(Honey, Kirchner, & VanRullen, 2008). One of theimages was a face, and the other one a vehicleVas inthe experiments reported here. They found a very strongface bias in that even when the images had the samecontrast, 70% of the saccades were made toward the faces.They then showed that at least some of this bias was stillpresent even when the image was completely phasescrambled in the Fourier domainVthat is, the imagesretained the same spatial frequency and orientationstructure as the original images. This result supports theidea that rudimentary and quickly accessible informationcould explain a part of the bias we observed. Interestingly,a similar bias could also explain the tendency of faces topop-out in multiple search arrays (Vanrullen, 2006). Theunderlying idea is that the visual system could use asimple heuristic characteristic of faces (for example, apattern of high energy on both horizontal high and verticallow spatial frequencies), can be computed very rapidly,and used to generate useful selectivity early on in visualprocessing.

Conclusions

We have shown using a Saccadic Choice Task thathumans can make very rapid saccades toward imagescontaining faces. The earliest reliable saccades can beseen from as little as 100 to 110 ms after stimulus onset.While early face-selective electrical activity has beenreported in a number of previous studies, this is the firsttime that it has been clear that such early selectiveresponses could have a clear impact on behavior. Thestudy also shows that the Saccadic Choice Task providesan experimental tool for studying very early processing in

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 13

the visual system, in a time window previously onlyaccessible using electrophysiological techniques.

Acknowledgments

H. K. was supported by a European grant “DecisionsIn Motion” and by the ANR project “Hearing in Time”.S. M. C. is supported by a grant from the DelegationGenerale pour l’Armement. The research was also financedby the CNRS and by the ANR “Natstats” Project.

Commercial relationships: none.Corresponding author: Simon Thorpe.Email: simon.thorpe@cerco.ups-tlse.fr.Address: Centre de Recherche Cerveau and Cognition,CNRS, University Toulouse 3, Toulouse, France.

References

Bentin, S., Allison, T., Puce, A., Perez, E., & Mccarthy, G.(1996). Electrophysiological studies of face perceptionin humans. Journal of Cognitive Neurosciences, 8,551–565.

Bindemann, M., Burton, A. M., Hooge, I. T., Jenkins, R.,& de Haan, E. H. (2005). Faces retain attention.Psychonomic Bulletin & Review, 12, 1048–1053.[PubMed]

Birmingham, E., Bischof, W. F., & Kingstone, A. (2008).Gaze selection in complex social scenes. VisualCognition, 16, 341–355.

Braeutigam, S., Bailey, A. J., & Swithenby, S. J. (2001).Task-dependent early latency (30–60 ms) visualprocessing of human faces and other objects. Neuro-report, 12, 1531–1536. [PubMed]

Brown, V., Huey, D., & Findlay, J. M. (1997). Facedetection in peripheral vision: Do faces pop out?Perception, 26, 1555–1570. [PubMed]

Buswell, G. T. (1935). How people look at pictures: Astudy of the psychology of perception in art. Chicago:University of Chicago Press.

Butler, S., Gilchrist, I. D., Burt, D. M., Perrett, D. I.,Jones, E., & Harvey, M. (2005). Are the perceptualbiases found in chimeric face processing reflected ineye-movement patterns? Neuropsychologia, 43, 52–59.[PubMed] [Article]

Cerf, M., Harel, J., Einhauser, W., & Koch, C. (2008).Predicting human gaze using low-level saliencycombined with face detection. Advances in neuralinformation processing systems (vol. 20, pp. 241–248).Cambridge, MA: MIT Press.

Cohen, J. D., & Tong, F. (2001). Neuroscience. The faceof controversy. Science, 293, 2405–2407. [PubMed]

Crouzet, S., Kirchner, H., & Thorpe, S. J. (2008).Saccading towards faces in 100 ms: What’s thesecret? Perception, 37(Supplement), 119–120.

Dakin, S. C., & Watt, R. J. (2009). Biological “bar codes”in human faces. Journal of Vision, 9(4):2, 1–10,http://journalofvision.org/9/4/2/, doi:10.1167/9.4.2.[PubMed] [Article]

Delorme, A., & Thorpe, S. J. (2001). Face identificationusing one spike per neuron: Resistance to imagedegradations. Neural Networks, 14, 795–803.[PubMed] [Article]

Downing, P. E., Jiang, Y., Shuman, M., & Kanwisher, N.(2001). A cortical area selective for visual processingof the human body. Science, 293, 2470–2473.[PubMed]

Drewes, J., Trommershaeuser, J., & Gegenfurtner, K. R.(2009). The effect of context on rapid animaldetection [Abstract]. Journal of Vision, 9(8):1177,1177a, http:/ / journalofvision.org/9/8/1177/,doi:10.1167/9.8.1177.

Eimer, M., & Holmes, A. (2002). An ERP study on thetime course of emotional face processing. Neuro-report, 13, 427–431. [PubMed]

Einhauser, W., Rutishauser, U., & Koch, C. (2008). Task-demands can immediately reverse the effects ofsensory-driven saliency in complex visual stimuli.Journal of Vision, 8(2):2, 1–19, http://journalofvision.org/8/2/2/, doi:10.1167/8.2.2. [PubMed] [Article]

Epshtein, B., Lifshitz, I., & Ullman, S. (2008). Imageinterpretation by a single bottom-up top-down cycle.Proceedings of the National Academy of Sciences ofthe United States of America, 105, 14298–14303.[PubMed] [Article]

Fabre-Thorpe, M., Delorme, A., Marlot, C., & Thorpe, S.(2001). A limit to the speed of processing in ultra-rapid visual categorization of novel natural scenes.Journal of Cognitive Neuroscience, 13, 171–180.[PubMed]

Farah, M. J., Wilson, K. D., Drain, M., & Tanaka, J. N.(1998). What is “special” about face perception?Psychological Review, 10, 482–498. [PubMed]

Fischer, B., & Weber, H. (1993). Express saccades andvisual attention. Behavioral and Brain Sciences, 16,553–610.

Fletcher-Watson, S., Findlay, J. M., Leekam, S. R., &Benson, V. (2008). Rapid detection of person infor-mation in a naturalistic scene. Perception, 37, 571–583.[PubMed]

Freiwald, W. A., Tsao, D. Y., & Livingstone, M. S.(2009). A face feature space in the macaque temporal

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 14

lobe. Nature Neuroscience, 12, 1187–1196. [PubMed][Article]

George, N., Jemel, B., Fiori, N., & Renault, B. (1997).Face and shape repetition effects in humans: A spatio-temporal ERP study. Neuroreport, 8, 1417–1423.[PubMed]

Gilchrist, I. D., & Proske, H. (2006). Anti-saccades awayfrom faces: Evidence for an influence of high-levelvisual processes on saccade programming. Exper-imental Brain Research, 173, 708–712. [PubMed]

Guyonneau, R., Kirchner, H., & Thorpe, S. J. (2006).Animals roll around the clock: The rotation invari-ance of ultrarapid visual processing. Journal ofVision, 6(10):1, 1008–1017, http://journalofvision.org/6/10/1/, doi:10.1167/6.10.1. [PubMed] [Article]

Haxby, J. V., Gobbini, M. I., Furey, M. L., Ishai, A.,Schouten, J. L., & Pietrini, P. (2001). Distributed andoverlapping representations of faces and objects inventral temporal cortex. Science, 293, 2425–2430.[PubMed]

Haxby, J. V., Hoffman, E. A., & Gobbini, M. I. (2000).The distributed human neural system for face percep-tion. Trends in Cognitive Sciences, 4, 223–233.[PubMed]

Hemond, C. C., Kanwisher, N. G., & Op de Beeck, H. P.(2007). A preference for contralateral stimuli inhuman object- and face-selective cortex. PLoS One,2, e574. [PubMed] [Article]

Henderson, J. M. (2003). Human gaze control during real-world scene perception. Trends in Cognitive Sciences,7, 498–504. [PubMed] [Article]

Hershler, O., & Hochstein, S. (2005). At first sight: Ahigh-level pop out effect for faces. Vision Research,45, 1707–1724. [PubMed] [Article]

Hershler, O., & Hochstein, S. (2006). With a careful look:Still no low-level confound to face pop-out. VisionResearch, 46, 3028–3035. [PubMed] [Article]

Hjelmas, E., & Low, B. K. (2001). Face detection: Asurvey. Computer Vision and Image Understanding,83, 236–274.

Honey, C., Kirchner, H., & VanRullen, R. (2008). Facesin the cloud: Fourier power spectrum biases ultrarapidface detection. Journal of Vision, 8(12):9, 1–13,http://journalofvision.org/8/12/9/, doi:10.1167/8.12.9.[PubMed] [Article]

Hung, C. P., Kreiman, G., Poggio, T., & Dicarlo, J. J.(2005). Fast readout of object identity from macaqueinferior temporal cortex. Science, 310, 863–866.[PubMed]

Ishai, A., Ungerleider, L. G., Martin, A., Schouten, J. L.,& Haxby, J. V. (1999). Distributed representation of

objects in the human ventral visual pathway. Pro-ceedings of the National Academy of Sciences of theUnited States of America, 96, 9379–9384. [PubMed][Article]

Itti, L., & Koch, C. (2000). A saliency-based searchmechanism for overt and covert shifts of visualattention. Vision Research, 40, 1489–1506. [PubMed][Article]

Jacques, C., & Rossion, B. (2009). The initial representa-tion of individual faces in the right occipito-temporalcortex is holistic: Electrophysiological evidence fromthe composite face illusion. Journal of Vision, 9(6):8,1–16, http://journalofvision.org/9/6/8/, doi:10.1167/9.6.8. [PubMed] [Article]

Jeffreys, D. A. (1989). A face-responsive potentialrecorded from the human scalp. Experimental BrainResearch, 78, 193–202. [PubMed]

Johnson, J. S., & Olshausen, B. A. (2003). Time course ofneural signatures of object recognition. Journal of Vision,3(7):4, 499–512, http://journalofvision.org/3/7/4/,doi:10.1167/3.7.4. [PubMed] [Article]

Johnson, J. S., & Olshausen, B. A. (2005). The earliestEEG signatures of object recognition in a cued-target task are postsensory. Journal of Vision, 5(4):2,299–312, http://journalofvision.org/5/4/2/, doi:10.1167/5.4.2. [PubMed] [Article]

Johnson, M. H. (2005). Subcortical face processing.Nature Reviews. Neuroscience, 6, 787–798.[PubMed]

Johnson, M. H., Dziurawiec, S., Ellis, H., & Morton, J.(1991). Newborns’ preferential tracking of face-likestimuli and its subsequent decline. Cognition, 40,1–19. [PubMed] [Article]

Johnson, M. H., & Mareschal, D. (2001). Cognitive andperceptual development during infancy. CurrentOpinion in Neurobiology, 11, 213–218. [PubMed][Article]

Kano, F., & Tomonaga, M. (2009). How chimpanzeeslook at pictures: A comparative eye-tracking study.Proceedings of the Royal Society B: BiologicalSciences, 276, 1949–1955. [PubMed]

Kanwisher, N. (2000). Domain specificity in face percep-tion. Nature Neuroscience, 3, 759–763. [PubMed]

Kanwisher, N., McDermott, J., & Chun, M. M. (1997).The fusiform face area: A module in human extras-triate cortex specialized for face perception. Journalof Neuroscience, 17, 4302–4311. [PubMed] [Article]

Kiani, R., Esteky, H., & Tanaka, K. (2005). Differences inonset latency of macaque inferotemporal neuralresponses to primate and non-primate faces. Journalof Neurophysiology, 94, 1587–1596. [PubMed][Article]

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 15

Kirchner, H., & Thorpe, S. J. (2006). Ultra-rapid objectdetection with saccadic eye movements: Visualprocessing speed revisited. Vision Research, 46,1762–1776. [PubMed] [Article]

Langton, S. R., Law, A. S., Burton, A.M., & Schweinberger,S. R. (2008). Attention capture by faces. Cognition,107, 330–342. [PubMed] [Article]

Liu, H., Agam, Y., Madsen, J. R., & Kreiman, G. (2009).Timing, timing, timing: Fast decoding of objectinformation from intracranial field potentials inhuman visual cortex. Neuron, 62, 281–290. [PubMed]

Liu, J., Harris, A., & Kanwisher, N. (2002). Stages ofprocessing in face perception: An MEG study. NatureNeuroscience, 5, 910–916. [PubMed]

Masquelier, T., & Thorpe, S. J. (2007). Unsupervisedlearning of visual features through spike timingdependent plasticity. PLoS Computational Biology,3, e31. [PubMed] [Article]

McCarthy, G., Puce, A., Belger, A., & Allison, T. (1999).Electrophysiological studies of human face percep-tion. II: Response properties of face-specific poten-tials generated in occipitotemporal cortex. CerebralCortex, 9, 431–444. [PubMed]

Mormann, F., Kornblith, S., Quiroga, R. Q., Kraskov, A.,Cerf, M., Fried, I., et al. (2008). Latency andselectivity of single neurons indicate hierarchicalprocessing in the humanmedial temporal lobe. Journalof Neuroscience, 28, 8865–8872. [PubMed] [Article]

Mouchetant-Rostaing, Y., & Giard, M. H. (2003). Electro-physiological correlates of age and gender perceptionon human faces. Journal of Cognitive Neuroscience,15, 900–910. [PubMed]

Mouchetant-Rostaing, Y., Giard, M. H., Bentin, S.,Aguera, P. E., & Pernier, J. (2000). Neurophysiolog-ical correlates of face gender processing in humans.European Journal of Neuroscience, 12, 303–310.[PubMed]

New, J., Cosmides, L., & Tooby, J. (2007). Category-specific attention for animals reflects ancestral prior-ities, not expertise. Proceedings of the NationalAcademy of Sciences, 104, 16598–16603. [PubMed][Article]

Ohman, A., Carlsson, K., Lundqvist, D., & Ingvar, M.(2007). On the unconscious subcortical origin ofhuman fear. Physiology Behavior, 92, 180–185.[PubMed] [Article]

Oram, M. W., & Perrett, D. I. (1992). Time course ofneural responses discriminating different views of theface and head. Journal of Neurophysiology, 68, 70–84.[PubMed]

Parkhurst, D., Law, K., & Niebur, E. (2002). Modeling therole of salience in the allocation of overt visual

attention. Vision Research, 42, 107–123. [PubMed][Article]

Peters, R. J., Iyer, A., Itti, L., & Koch, C. (2005).Components of bottom-up gaze allocation in naturalimages. Vision Research, 45, 2397–2416. [PubMed][Article]

Puce, A., Allison, T., Gore, J. C., & McCarthy, G. (1995).Face-sensitive regions in human extrastriate cortexstudied by functional MRI. Journal of Neurophysiol-ogy, 74, 1192–1199. [PubMed]

Qiu, F. T., Sugihara, T., & von der Heydt, R. (2007).Figure–ground mechanisms provide structure for selec-tive attention. Nature Neurosciences, 10, 1492–1499.[PubMed] [Article]

Ro, T., Russell, C., & Lavie, N. (2001). Changing faces: Adetection advantage in the flicker paradigm. Psycho-logical Science, 12, 94–99. [PubMed]

Roelfsema, P. R., Tolboom, M., & Khayat, P. S. (2007).Different processing phases for features, figures, andselective attention in the primary visual cortex.Neuron, 56, 785–792. [PubMed]

Rossion, B., & Jacques, C. (2008). Does physicalinterstimulus variance account for early electrophy-siological face sensitive responses in the humanbrain? Ten lessons on the N170. Neuroimage, 39,1959–1979. [PubMed] [Article]

Rousselet, G. A., Mace, M. J., & Fabre-Thorpe, M. (2003).Is it an animal? Is it a human face? Fast processing inupright and inverted natural scenes. Journal of Vision,3(6):5, 440–455, http://journalofvision.org/3/6/5/,doi:10.1167/3.6.5. [PubMed] [Article]

Rousselet, G. A., Mace, M. J., & Fabre-Thorpe, M.(2004). Animal and human faces in natural scenes:How specific to human faces is the N170 ERPcomponent? Journal of Vision, 4(1):2, 13–21, http://journalofvision.org/4/1/2/, doi:10.1167/4.1.2.[PubMed] [Article]

Rousselet, G. A., Mace, M. J., Thorpe, S. J., & Fabre-Thorpe, M. (2007). Limits of event-related potentialdifferences in tracking object processing speed.Journal of Cognitive Neuroscience, 19, 1241–1258.[PubMed]

Schroeder, C. E., Mehta, A. D., & Givre, S. J. (1998). Aspatiotemporal profile of visual system activationrevealed by current source density analysis in theawake macaque. Cerebral Cortex, 8, 575–592.[PubMed]

Sergent, J., Ohta, S., & MacDonald, B. (1992). Functionalneuroanatomy of face and object processing. Apositron emission tomography study. Brain: A Jour-nal of Neurology, 115, 15–36. [PubMed]

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 16

Serre, T., Oliva, A., & Poggio, T. (2007). A feedforwardarchitecture accounts for rapid categorization. Pro-ceedings of the National Academy of Sciences of theUnited States of America, 104, 6424–6429. [PubMed][Article]

Sinha, P., Balas, B., & Ostrovsky, Y. (2007). Discoveringfaces in infancy [Abstract]. Journal of Vision, 7(9):569,569a, http://journalofvision.org/7/9/569/, doi:10.1167/7.9.569.

Tarr, M. J., & Gauthier, I. (2000). FFA: A flexiblefusiform area for subordinate-level visual processingautomatized by expertise. Nature Neuroscience, 3,764–769. [PubMed]

Tatler, B. W., Baddeley, R. J., & Gilchrist, I. D. (2005).Visual correlates of fixation selection: Effects of scaleand time. Vision Research, 45, 643–659. [PubMed][Article]

Taubert, J. (2009). Chimpanzee faces are “special” tohumans. Perception, 38, 343–356. [PubMed]

Theeuwes, J., & Van der Stigchel, S. (2006). Facescapture attention: Evidence from inhibition of return.Visual Cognition, 13, 657–665.

Thorpe, S., Fize, D., & Marlot, C. (1996). Speed ofprocessing in the human visual system. Nature, 381,520–522. [PubMed]

Thorpe, S., & Imbert, M. (1989). Biological con-straints on connectionist modelling. In R. Pfeifer,Z. Schreter, F. Fogelman-Soulie, & L. Steels (Eds.),Connectionism in perspective (pp. 63–93). Amsterdam,The Netherlands: Elsevier.

Treue, S. (2001). Neural correlates of attention in primatevisual cortex. Trends in Neurosciences, 24, 295–300.[PubMed] [Article]

Tsao, D. Y., & Livingstone, M. S. (2008). Mechanisms offace perception. Annual Review of Neuroscience, 31,411–437. [PubMed] [Article]

Vanrullen, R. (2006). On second glance: Still no high-level pop-out effect for faces. Vision Research, 46,3017–3027. [PubMed] [Article]

VanRullen, R., Gautrais, J., Delorme, A., & Thorpe, S.(1998). Face processing using one spike per neurone.BioSystems, 48, 229–239. [PubMed]

VanRullen, R., & Thorpe, S. J. (2001a). Is it a bird? Is it aplane? Ultra-rapid visual categorisation of natural andartifactual objects. Perception, 30, 655–668.[PubMed]

VanRullen, R., & Thorpe, S. J. (2001b). The time courseof visual processing: From early perception todecision-making. Journal of Cognitive Neuroscience,13, 454–461. [PubMed]

VanRullen, R., & Thorpe, S. J. (2002). Surfing a spikewave down the ventral stream. Vision Research, 42,2593–2615. [PubMed] [Article]

Viola, P., & Jones, M. J. (2004). Robust real-time facedetection. International Journal of Computer Vision,57, 137–154.

Vuilleumier, P. (2000). Faces call for attention: Evidencefrom patients with visual extinction. Neuropsycholo-gia, 38, 693–700. [PubMed] [Article]

Yarbus, A. F. (1967). Eye movements and vision. NewYork: Plenum Press.

Journal of Vision (2010) 10(4):16, 1–17 Crouzet, Kirchner, & Thorpe 17